Antiretroviral therapy use was assessed via self-report at every visit in the WIHS and MACS and a binary variable was used to denote any ART use at every study visit. Plasma HIV viral load was measured using standard laboratory techniques and was classified as suppressed if the viral load was less than 200 copies/mL and unsuppressed if 200 copies/mL or greater. Other covariates included BMI and high-density lipoprotein cholesterol.We used frequencies and percentages for categorical variables and medians and interquartile ranges for continuous variables to describe characteristics of participants in the WIHS and MACS at the index visit. Next, we conducted time-dependent Cox regression models to estimate the association between self-reported frequency of marijuana use and incident type 2 diabetes separately for WIHS and MACS for the combined group and separately by HIV status. The primary predictor variable in these models was the time-dependent self-reported frequency of marijuana use , with none designated as the reference group. The outcome variable was incident type 2 diabetes ascertained over the follow-up period. Time to incident type 2 diabetes was calculated as the number of years from the index visit until the visit when type 2 diabetes was ascertained or until administrative censoring . We estimated both unadjusted and fully adjusted hazard ratios,indoor garden table with 95% CIs from the Cox regression models. Fully adjusted models included age, race/ethnicity, educational attainment , study center, study enrollment cohort, family history of diabetes and time-dependent covariates including alcohol use, smoking status, stimulant use, high blood pressure, and BMI.
The models for HIV-positive participants were additionally adjusted for time-dependent ART use and viral suppression status. We explored potential effect modification by BMI, race/ethnicity status, educational attainment, and smoking status in all models by testing multiplicative interaction terms with marijuana use variable. Prior studies have found that cannabis use is associated with lower fasting glucose, Hemoglobin A1C and BMI . Therefore, in secondary analysis, we used generalized estimating equations to assess the longitudinal associations between self-reported frequency of marijuana use and BMI, fasting blood glucose and hemoglobin A1C using the same follow-up period as the primary analysis. We accounted for the dependency between repeated measurements of the outcomes by robust estimation of error variances and specifying an unstructured correlation structure. Missing data for the primary predictor and covariates were addressed by means of multiple imputation using chained equations . Ten imputed data sets where generated for missing time-stable and time-dependent covariates; Cox regression models were conducted on each imputed data set and the hazard ratios were pooled . All statistical analyses were performed using SAS version 9.4 . The final analysis data set included 3,578 participants in the WIHS and 2,682 in the MACS , and their characteristics at the index visit are displayed in Table 1. The median age in the WIHS was 39 years and 43 years in the MACS. Most men in the MACS were non-Hispanic white , followed by non-Hispanic black . In contrast, non-Hispanic black and Hispanic women predominated in the WIHS. Eighteen percent of women in the WIHS self reported any marijuana use in the past six months, whereas 36% self-reported any marijuana use in the MACS.
Women in the WIHS accumulated 30,800 person-years of follow-up , during which 452 incident cases of type 2 diabetes were ascertained, representing an unadjusted incident rate of type 2 diabetes of 1.46 per 100 person-years among all women in the WIHS. Conversely, the men in the MACS accumulated 31,281 person-years of follow-up , with 326 incident cases of type 2 diabetes ascertained , and an unadjusted incident rate of type 2 diabetes of 1.04 per 100 person-years among all men in the MACS. In the unadjusted models for the association between self-reported frequency of marijuana use and incident type 2 diabetes, among all women in the WIHS, we found a lower risk of type 2 diabetes in daily users compared with nonusers , with the confidence intervals around the hazard ratio estimate excluding the null value. However, adjusting for BMI and other covariates, attenuated the association and increased the width of the confidence interval to include the null value . Furthermore, in the models for all women in the WIHS and separately by HIV-status, the point estimates of the fully adjusted hazard ratios for self-reported monthly/less and weekly/less levels of marijuana use , suggest a very small increased risk of type 2 diabetes compared to non-use. The point estimates for self-reported daily marijuana use indicated a reduced risk of type 2 diabetes compared to non-use. However, the confidence intervals around all the fully adjusted hazard ratios were relatively wide – spanning both the reduced, null and increased risk range . In the models for all men in the MACS and separately by HIV-status, the point estimates for the fully adjusted hazard ratios for most self-reported frequency levels marijuana use, collectively suggest a reduced risk of type 2 diabetes compared to non-use.
Like the WIHS, the confidence intervals around the fully adjusted hazard ratios were wide, spanning both the reduced, null and increased risk range . The fully adjusted models with covariate estimates for all models in the WIHS and MACS are included in the Supplementary Material 1. Tests of effect modification by BMI, race/ethnicity, educational attainment, and smoking status were not statistically significant . In secondary analysis, in both the WIHS and MACS cohort, we found a pattern of statistically significant dose-dependent associations between increasing self-reported frequency of marijuana use and lower BMI. Specifically, as self-reported frequency of marijuana use increased, BMI decreased, with confidence intervals around all point estimates excluding the null value. Results of models evaluating the longitudinal associations between self-reported frequency of marijuana use and BMI are included in the Supplemental Material 2. Additionally, there was a pattern of lower fasting blood glucose and hemoglobin A1C with increasing self-reported frequency of marijuana use in all men in the MACS cohort, as well as in most subgroup analysis ,grow rack with most estimates excluding the null value . However, in the WIHS cohort, although there was a consistent pattern of lower hemoglobin A1C associated with self-reported frequency of marijuana use there was a consistent pattern of higher fasting blood glucose in self-reported daily marijuana users compared to nonusers in all women in the WIHS and in HIV-positive and HIV-negative women, although the confidence intervals around the point estimates, all included the null value. Results for association of self reported marijuana use and BMI, hemoglobin A1C and fasting glucose are included in the supplemental material 2. We conducted sensitivity analysis to explore whether participants with imminent type 2 diabetes ceased or reduced marijuana use because of worsening health symptoms , as has been observed in studies involving alcohol use . Therefore, we lagged the marijuana use variable by one time point and re-ran all models. The pattern of results remained the same. In the MACS, although all the hazard ratios for self reported marijuana use frequency in relation to incident type 2 diabetes indicate a reduced risk, they were not statistically significant. Further, the pattern of findings in the WIHS were similar and not statistically significant. Results for the lagged analysis of the association of self-reported frequency of marijuana use and incident type 2 diabetes are included in the Supplemental Material 3. As with BMI, high blood pressure could potentially lie on the causal pathway between marijuana use and type 2 diabetes. Thus, we re-ran analysis with and without high blood pressure, but the results remained the same. This analysis examined associations of self-reported frequency of marijuana use and incident type 2 diabetes in women and men living with and at risk for HIV in the WIHS and MACS cohorts respectively followed prospectively for nearly 20 years.
Among men in the MACS, we found that self-reported frequency of marijuana use – including daily use – was associated with a reduced risk of type 2 diabetes, among all men and in those living with and at risk for HIV, although these associations were not statistically significant. Among women in the WIHS, daily marijuana use was associated with a reduced risk of type 2 diabetes, but there was a pattern of increased type 2 diabetes risk in women with less than daily marijuana use among all women and in those living with and at risk for HIV. Albeit, these associations were not statistically significant. To our knowledge, this is one of the largest longitudinal follow-up studies examining the relationship between marijuana use and incident type 2 diabetes in men and women living with or at risk for HIV. Data for our analysis came from two large, longitudinal cohorts with multiple observations over a long follow-up among persons living with or at-risk for HIV in midlife, when risks for incident type 2 diabetes are increased. Prior studies of the association between marijuana use and prevalence of diabetes have suggested lower odds of diabetes in subjects who use marijuana . One meta-analyses of eight different cross-sectional studies involving nationally representative surveys of the U.S population found an overall reduced odds of diabetes among subject who use marijuana . Another recent cross-sectional U.S. population-based study using the National Epidemiological Survey on Alcohol and Related Conditions that reported a statistically significant reduced odds of type 2 diabetes among lifetime and past 12‐ month marijuana use compared with never use . However, these studies utilized cross-sectional data, and thus evaluated associations between marijuana use and prevalence diabetes. Furthermore, findings from this analysis are consistent with those from two other longitudinal studies that assessed associations between marijuana use and risk of diabetes conducted in the general population. Bancks et al followed up healthy men and women in the CARDIA Study over 18 years and reported hazard ratios indicating increased risk of diabetes in relation to greater lifetime frequency of marijuana use, but with wide confidence intervals around the hazard ratios that included the null value. Similarly, in a population-based cohort study of Swedish men and women followed up for more than 8 years, the authors found an adjusted protective odds ratio, between lifetime marijuana use and diabetes risk, with wide confidence intervals around this point estimates that included the null estimates . The finding from our study and these two longitudinal studies parallel the association between marijuana use and type 2 diabetes, conducted in different populations likely underscores the power of long term longitudinal assessments – when compared to cross-sectional studies, particularly when trying to measure incidence of a chronic disease linked with marijuana use over the life span. The hazard ratios from the fully adjusted analysis for men in the MACS cohort, indicated a reduced risk of type 2 diabetes across all self-reported frequency of marijuana use . Although, the confidence intervals around the adjust hazard ratios for the associations between self-reported frequency of marijuana use and incident type 2 diabetes were wide – including both the reduced, null and increase risk range – they leaned toward the reduced risk range. However, in the WIHS, the hazard ratios from the fully adjusted analysis indicated a reduced risk of type 2 diabetes only among daily users, with other user categories leaning toward an increased risk of type 2 diabetes. These findings may point to potential sex differences in the association between marijuana use and type 2 diabetes and metabolic health. This will be consistent with other studies showing sex differences in the relationship between marijuana use and health . As such, assessing sex differences of marijuana use on metabolic health is warranted, particularly in the context of HIV. The mechanisms of a proposed marijuana and type 2 diabetes association, suggest that THC stimulates appetite through activation of CB1 , and thus may play a role in eating behaviors among persons who use marijuana. Yet, the preponderance of studies documents better metabolic indicators in marijuana users compared with nonusers including higher high-density lipoprotein cholesterol levels; lower levels of fasting glucose, triglycerides, fasting insulin , and Homeostasis Model Assessments of Insulin Resistance ; smaller waist circumferences ; lower BMI ; and lower odds of metabolic syndrome . Indeed, in further analysis of WIHS and MACS data here, we found a statistically significant dose-dependent association between self-reported frequency of marijuana use and lower BMI. Results are included in the Supplemental Material 4.